An innovative model of localization error of ground targets by using drones

This paper describes a model for evaluating the change detection accuracy of an identified target by using a visible band camera installed on-board a drone. An automatic detection process is used to recognize a known target in the surrounding environment, and a geometrical model for target positioning is developed according to two different camera orientations, i.e. downward and forward pointing camera. The uncertainties on drone position, sensor pointing direction, and target position in the obtained image are modelled and estimated. The proposed model for target localization accuracy is developed adopting a parametric strategy to be adapted to different cameras and drones. To assess the performance of the model, flight tests are performed through an advanced commercial drone equipped with a high-resolution camera. For the reported use case, the results show that the Root Mean Square Error on each target position component evaluated by the proposed model is <0.25 m, in agreement with the results achieved by the flight tests. The proposal shows itself suitable to support management and maintenance strategies for buildings and infrastructures. Specifically, railway inspection can be improved by using drones both to detect and localize a target-of-interest, such as balise, light signal, buffer stop, and to evaluate the target position accuracy starting from the used platform and camera as well as from geometry definition.